Short-Term Load Forecasting Based on VMD and PSO Optimized Deep Belief Network

In order to improve accuracy of short-term load forecasting, original historical load sequence is decomposed into several characteristic model functions with variational mode decomposition (VMD). Load forecasting models are developed after analyzing characteristics of each model function. Selecting effective input variables is technical measures to improve load forecasting accuracy. In this paper, mutual information is adopted to calculate correlation between influence factors and output variables, and then an input variable set is selected. It is difficult for traditional load forecasting model based on neural network to train multi-layer network, thus affecting its prediction accuracy. Deep belief network (DBN) uses a non-supervised greedy layer-by-layer training algorithm to construct a multi-hidden layer sensor structure with excellent performance in regression and forecasting analysis, and becomes a research hotspot in deep learning field. This paper uses DBN algorithm to establish a forecasting model for each model function to improve prediction accuracy. DBN connection weight is optimized with particle swarm optimization algorithm to avoid local optimal solution due to random initialization, thus enhancing DBN forecasting performance. Finally, case test shows effectiveness of the proposed model. © 2018, Power System Technology Press. All right reserved.